1. Field of the Invention
The present invention relates to electrostatic switches, and more particularly, to laterally-bending electrostatic switches.
2. Description of the Related Art
The invention of the bipolar transistor in 1945 came in a time of analog electronics. At that time, dynamic signal amplification was a necessity. The advent of digital logics has eliminated this requirement, yet transistors are still used as switches in integrated circuits despite their complicated doped structures, single crystal material requirements, intricate contacting to metal interconnects, sensitivity to radiation and limited active temperature range (below 150 C).
In spite of this, performance in the semiconductor industry has increased exponentially in the last 40 years. This was accomplished by scaling down a transistor's size, increasing die size and enhancing the interconnect efficiency. The scaling of transistors reduces the intrinsic switching delay and dissipation. Scaling of interconnects however results in larger interconnect lengths which leads to an increase in the latency or response time and energy dissipation as compared to that in transistors. For submicron technology “RC” delays from interconnects becomes a dominant factor. It was expected that a change to copper interconnects (lower R) and low-k dielectrics (lower C) would allow the exponential increase in performance to continue through this decade.
While copper interconnects have been successfully integrated to the fabrication process the same cannot be said for low-k dielectrics, which seems to indicate the end for device scaling, as we know it. The current situation urges the search for alternatives for current 2-D silicon technology.
A possible solution is 3-D integration. Replacing the long interconnects in 2-D structures by short vertical ones would enhance performance. This would also allow the integration of different technologies and reduce the size of the final package.
Three main 3-D integration techniques have been proposed: multichip stacks, recrystallized silicon, and monolithic wafer level integration. An extraordinary challenge for the 3-D integration is how to go effectively from silicon to interconnect to silicon again.
As the name indicates, multichip stacks consist of interconnecting fully processed chips together by lining up bond pads on each chip and solder them. One of the problems of this approach is the lack of precision alignment, which restricts the number and size of interconnects. The silicon between layers also creates thermal resistance and reduces heat dissipation. Meanwhile, the more layers the chip has the more space is lost to vias for interconnects, reducing the available space for active elements such as transistors.
In the recrystallized silicon approach, a layer of polycrystalline silicon is grown on top of the substrate and then partially recrystallized by heating it. This creates new silicon layers in which to create transistors and other elements. This has the advantage of a thinner silicon layer than in the multichip stack approach. However, high temperatures for recrystallization would damage circuits in the bottom layers. Also the carrier mobility on the polysilicon is lower compared to the one in single-crystal silicon, which slows performance.
In the monolithic wafer-level integration entire wafers are glued and then cut into single chips. This would reduce costs as compared to the multichip stack process because an entire wafer with many chips is handled at once. Fabricators will have similar problems on wafer alignment, bonding, silicon thinning (to reduce thermal resistance) and interconnect connection. The complexity and costs that wafer-scale integration implies could render the whole task impossible.
In addition, no high speed (<10 ns) non-volatile semiconductor memory elements have ever been realized. Some memory elements based on magnetic tunnel junctions, phase change materials, programmable metallization cells, nanotube arrays and ionic polymers have been designed, however, all require integration with metal-semiconductor circuits and none have demonstrated cost effective high-speed high-density operation.
Micromechanical electrostatic switches have been proposed for switching of RF signals. Some examples of such micromechanical electrostatic switches are shown and discussed in “Microelectromechanical systems (MEMS): fabrication, design and applications,” by J. Judy, Smart, Mater. Struct. 10 (2001) 1115-1134. The switching itself in a micromechanical electrostatic switch is usually on the order of microseconds.
An example is show in FIG. 1. The operation principle is simple: with a positive voltage on the source, a negative voltage on the gate will attract the beam and will form an electrical short between source and drain. The force between the gate and source can be approximated by the equation for a force between parallel capacitor plates:
  F  =            ɛ      0        ⁢    A    ⁢                  V        2                    d        2            where ε0 is the permittivity of vacuum, A is the gate area, V is the gate-source voltage and d is the distance between the gate electrode and source beam. The force of the electric field is counteracted by the spring force of the beam resulting in a threshold voltage for switching according to:
      V    th    =            2      3        ⁢    d    ⁢                            2          ⁢          kd                          3          ⁢                                          ⁢                      ɛ            0                    ⁢          A                    
Here Vth is the switching threshold voltage, d the thickness of the beam, k the effective spring constant of the beam and A the area of the gate. This spring constant k can be approximated by:
  k  =            bt      3        ⁢                  Y        2                    4        ⁢                  L          3                    where b is the width, t the thickness and L the length of the beam.
In FIG. 1, the cantilever structures were approximately 65 microns in length, and 30 microns wide. The thickness of the cantilevers is approximately 2 microns and the beam-to-gate spacing is about 1.5 microns. The threshold voltage is about 50 volts and the device will operate at between 2 and 3 MHz. More than 1 million switch cycles have been demonstrated. Above this the gold contacts started to deteriorate due to fusing of gold atoms. For wear free switching inert contact coatings are needed.
The gate area, and thus force, is much larger than the drain area. This in principle reduces the attraction by the drain electrode and allows amplification. However, this requires large devices. An alternative is to electrically isolate the contacting area from the gate area, forming a relay instead of a switch (see U.S. Pat. No. 6,152,839 to Zavracky relating to micromechanical switching devices). This however is very complicated to fabricate. Although the principle of switching is useful for digital logics, this design is not sufficiently fast, operates at very high voltages and is too large and complicated to produce for integrated circuit applications.
A sufficiently fast low voltage design based on a carbon nanotube (CNT) beam has been proposed very recently and is depicted in FIG. 2. This is a schematic picture of the theoretical model system consisting of a conducting carbon nanotube (CNT) placed on a terraced Si substrate. The terrace height is labeled h, and q denotes the excess charge on the tube. The CNT is connected to a source electrode (S), and the gate (G) and drain (D) electrodes are placed on the substrate beneath the CNT at lengths L and LG away from the terrace. The displacement x of the nanotube tip is measured towards the substrate. Typically, L is about 50-100 nm, h=5 nm.
This prior art theoretical proposal has several advantages over the switch in FIG. 1, namely small dimensions, fast switching (1 ns), low voltage operation (IV) and potential absence of sticking due to the inert nature of the nanotube. Non-volatility is provided by the adhesion properties of the beam to the drain: sticking based on VanderWaals forces can lead to sufficient hysteresis. However, because the drain attracts the CNT just like the gate does, the source-drain voltage must always be lower than the source-gate voltage, which means that amplification is not possible in the design of FIG. 2. In other words, the drain output of the CNT switch cannot control the gate of another CNT switch, precluding logic operation. In addition, the realization and fabrication of this CNT switch is complicated. Growth of nanotubes is still unpredictable, both in number, orientation, dimension, location, and conductance type (metallic or semiconducting). Forming a high yield integrated circuit with millions of these switches with all the same properties and subnanometer accurate CNT beam positioning seems unrealistic within this geometry. In addition, forming low resistance metal interconnect-CNT contacts and low resistance beam-drain contact is challenging: a clear disadvantage of switches that are based on material foreign to metallic interconnects.
Similarly, electrostatic switches have been proposed with either a vertically moving cantilever (FIG. 3(a)) or laterally moving cantilever (FIG. 3(b)). Here counter drain electrodes with a permanent voltage have been suggested to provide an attractive force to the cantilever, to realize opening of the switch besides using stress in the cantilever. Thus, applying a source-gate voltage performs closing, and the constant source-control voltage realizes opening. The disadvantage of this structure is that a separate voltage line is needed and that the source-control voltage depends on the source voltage. This source voltage can vary throughout a logic circuit. Moreover, a very large gate area is needed to provide a difference between source-gate attraction and source-drain attraction, precluding nanoscale scaling of the switches. Furthermore, no contact materials have been specified that can support operating without wear at the contact site.
What is needed is an easily manufacturable nanometer size-amplifying switch that can operate fast (<1 ns), in a wide temperature range (up to 1000 C), using low power and low voltage and is radiation hard and wear-free. In addition, non-volatility is needed to provide reconfigurable logics and embedded memory. As interconnects are an essential part of proposed technologies, an scheme that incorporates interconnects and switching devices in similar fabrication steps would represent important advantages regarding simplicity of fabrication-and as a consequence cost-, RC delay times, interconnection and 3-D expansion.
Considering the fact that high conductance metals such as copper are still the ideal choice for low RC constant interconnects, a non-volatile switch made from this same metal in a straightforward geometry facilitates fabrication. It is an objective of this invention to provide a volatile and non-volatile switch that uses patterned interconnect metal for its structure and electrostatics to provide mechanical movement for switching. Another objective is to form an electronic circuit with reconfigurable logic and non-volatile memory from these elements. It is a further objective to form a 3D layered structure consisting of three-dimensional interconnects in which different layers are linked through volatile or non-volatile switches.